Precise streamflow forecasting is essential for managing water resources, controlling flooding, planning irrigation, and guaranteeing agricultural sustainability. This study employs 38 years of monthly flow of the stream data from the Cholachagudda dam on the Malaprabha River, India, to evaluate and compare the performance of deep learning models, including CNN, RNN, LSTM, GRU, and a Transformer-based architecture. With the lowest RMSE (76.02), MAE, and quantile loss (20.72), the Transformer model performs better than alternative topologies, according to the dataIt overcomes the drawbacks of sequential models like LSTM and GRU by successfully identifying complicated temporal patterns and long-range relationships through its self-attention mechanism. This study highlights the challenges of predicting streamflow in non-linear hydrological systems, particularly for models limited to temporal data. By offering a thorough assessment, this study highlights the revolutionary potential of deep learning in hydrological forecasting and establishes the foundation for future developments in streamflow prediction techniques.